github-platform-watch
On-demand deep research on the GitHub platform — Actions, GHAS, Audit Log, Billing, Issues, Enterprise EMU/SAML, repository governance. Same workflow as copilot-deep-dive but a different topic surface. Most "what's the current state of for an enterprise / regulated org?" questions land here.
When to use
- A GitHub platform question that's not Copilot-specific (Copilot questions →
copilot-deep-dive).
- Comparing a feature's current state to the org's existing posture (e.g., "should we move from CodeQL default setup to advanced setup?").
- A net-new question that should become durable knowledge in
vault/facts/github/.
When NOT to use
Obsidian-first workflow (mandatory)
- Query the vault first via
vault-querier:
- Full-text search the question's key terms across
vault/facts/github/**, vault/research/github/**, vault/insights/**, and recent vault/digests/** for GitHub-tagged items (last 90 days).
- Backlink check on
[[github]] and the question's entities (e.g. [[github-actions]], [[emu]]).
- Triage findings:
- If the vault answers the question fully → return the existing answer with source citations (vault path + original source URLs from the fact's frontmatter). No new write.
- If partial → identify the gap. Web research targets only the gap.
- If empty → full web research.
- A gap means the vault has no note ≤90 days old answering the question.
- Web research (only on confirmed gaps):
- Use
source-fetcher (with prompt-injection-guard) on tier-1 sources: GitHub Docs, the GitHub blog changelog, GitHub's public roadmap, official GitHub security advisories.
- Extract claims via
claim-extractor.
- Verify load-bearing claims via
verify-claim (3-vote refute):
- Any claim destined for
vault/facts/github/ must pass verification.
- Claims from GitHub's own docs/changelog are exempt (the vendor IS the authority).
- Claims from secondary sources (analyst blogs, practitioner posts) get the full 3-vote treatment.
- Write the research note via
digest-writer (which delegates the file write to vault-writer.write_research):
- Path:
vault/research/github/YYYY-MM-DD-{slug}.md
- Frontmatter per
research.yml schema: topic from the taxonomy below, question, sources, findings_count, verified_claims.
- Body: TL;DR + Findings (with quoted anchors) + Sources (with credibility-tier badges).
- Stage promotable claims to
_inbox/github-platform-watch/:
- Any verified fact-typed claim →
_inbox/github-platform-watch/{timestamp}-{slug}.md with suggested_surface: facts and suggested_path: facts/github/{predicate}.md.
memory-curator decides on its next sweep.
Topic taxonomy
When writing research notes, use the topic frontmatter field consistently. For this skill, valid topics include:
github-actions — Actions runners, workflows, marketplace, OIDC, hardening.
ghas-codeql — CodeQL specifically.
ghas-dependabot — Dependabot config, alerts, security updates.
ghas-secret-scanning — push protection, partner patterns, custom patterns.
github-audit-log — admin audit, security log, IP allow list, etc.
github-billing — usage metrics, license assignment, cost.
github-enterprise — EMU, SAML, SCIM, runner groups, policies.
github-repo-governance — branch protection, CODEOWNERS, rulesets, allowed actions.
The research note's path will be vault/research/github/YYYY-MM-DD-{slug}.md regardless of topic — the topic frontmatter field is what vault-querier filters on later.
Compliance-relevant framing
Same as copilot-deep-dive. Every finding ties to SR 11-7 / FFIEC / OCC / SOX ITGC / NYDFS 500 / data residency where applicable; #general tag otherwise.
Composes with
Same as copilot-deep-dive. The two skills share the on-demand research pattern; only the topic scope differs.
Acceptance test (for step 11 done-criteria)
One live end-to-end research run produces a research note at vault/research/github/YYYY-MM-DD-{slug}.md with the same structure as copilot-deep-dive's acceptance.
1---2name: github-platform-watch3description: github-platform-watch4---56# github-platform-watch78On-demand deep research on the GitHub platform — Actions, GHAS, Audit Log, Billing, Issues, Enterprise EMU/SAML, repository governance. Same workflow as [`copilot-deep-dive`](../copilot-deep-dive/SKILL.md) but a different topic surface. Most "what's the current state of <GitHub platform feature> for an enterprise / regulated org?" questions land here.910## When to use1112- A GitHub platform question that's not Copilot-specific (Copilot questions → `copilot-deep-dive`).13- Comparing a feature's current state to the org's existing posture (e.g., "should we move from CodeQL default setup to advanced setup?").14- A net-new question that should become durable knowledge in `vault/facts/github/`.1516## When NOT to use1718- Copilot questions → `copilot-deep-dive`.19- Specific repo audits → [`repo-golden-path-scorer`](../repo-golden-path-scorer/SKILL.md).20- Workflow hardening for a specific YAML file → [`actions-workflow-hardener`](../actions-workflow-hardener/SKILL.md).2122## Obsidian-first workflow (mandatory)23241. **Query the vault first** via `vault-querier`:25 - Full-text search the question's key terms across `vault/facts/github/**`, `vault/research/github/**`, `vault/insights/**`, and recent `vault/digests/**` for GitHub-tagged items (last 90 days).26 - Backlink check on `[[github]]` and the question's entities (e.g. `[[github-actions]]`, `[[emu]]`).272. **Triage findings**:28 - If the vault answers the question fully → return the existing answer with source citations (vault path + original source URLs from the fact's frontmatter). No new write.29 - If partial → identify the **gap**. Web research targets only the gap.30 - If empty → full web research.31 - A **gap** means the vault has no note ≤90 days old answering the question.323. **Web research** (only on confirmed gaps):33 - Use `source-fetcher` (with `prompt-injection-guard`) on tier-1 sources: GitHub Docs, the GitHub blog changelog, GitHub's public roadmap, official GitHub security advisories.34 - Extract claims via `claim-extractor`.354. **Verify load-bearing claims** via `verify-claim` (3-vote refute):36 - Any claim destined for `vault/facts/github/` must pass verification.37 - Claims from GitHub's own docs/changelog are exempt (the vendor IS the authority).38 - Claims from secondary sources (analyst blogs, practitioner posts) get the full 3-vote treatment.395. **Write the research note** via `digest-writer` (which delegates the file write to `vault-writer.write_research`):40 - Path: `vault/research/github/YYYY-MM-DD-{slug}.md`41 - Frontmatter per `research.yml` schema: `topic` from the taxonomy below, `question`, `sources`, `findings_count`, `verified_claims`.42 - Body: TL;DR + Findings (with quoted anchors) + Sources (with credibility-tier badges).436. **Stage promotable claims** to `_inbox/github-platform-watch/`:44 - Any verified fact-typed claim → `_inbox/github-platform-watch/{timestamp}-{slug}.md` with `suggested_surface: facts` and `suggested_path: facts/github/{predicate}.md`.45 - `memory-curator` decides on its next sweep.4647## Topic taxonomy4849When writing research notes, use the `topic` frontmatter field consistently. For this skill, valid topics include:5051- `github-actions` — Actions runners, workflows, marketplace, OIDC, hardening.52- `ghas-codeql` — CodeQL specifically.53- `ghas-dependabot` — Dependabot config, alerts, security updates.54- `ghas-secret-scanning` — push protection, partner patterns, custom patterns.55- `github-audit-log` — admin audit, security log, IP allow list, etc.56- `github-billing` — usage metrics, license assignment, cost.57- `github-enterprise` — EMU, SAML, SCIM, runner groups, policies.58- `github-repo-governance` — branch protection, CODEOWNERS, rulesets, allowed actions.5960The research note's path will be `vault/research/github/YYYY-MM-DD-{slug}.md` regardless of topic — the topic frontmatter field is what `vault-querier` filters on later.6162## Compliance-relevant framing6364Same as `copilot-deep-dive`. Every finding ties to SR 11-7 / FFIEC / OCC / SOX ITGC / NYDFS 500 / data residency where applicable; `#general` tag otherwise.6566## Composes with6768Same as `copilot-deep-dive`. The two skills share the on-demand research pattern; only the topic scope differs.6970## Acceptance test (for step 11 done-criteria)7172One live end-to-end research run produces a research note at `vault/research/github/YYYY-MM-DD-{slug}.md` with the same structure as `copilot-deep-dive`'s acceptance.